Security Implications of Deepfakes in Face Authentication

Abstract

Deepfakes are media generated by deep learning and are nearly indistinguishable from real content to humans. Deepfakes have seen a significant surge in popularity in recent years. There have been numerous papers discussing their effectiveness in deceiving people. What's equally, if not more concerning, is the potential vulnerability of facial and voice recognition systems to deepfakes. The misuse of deepfakes to spoof automated facial recognition systems can threaten various aspects of our lives, including financial security and access to secure locations. This issue remains largely unexplored. Thus, this paper investigates the technical feasibility of a spoofing attack on facial recognition. Firstly, we perform a threat analysis to understand what facial recognition use cases allow the execution of deepfake spoofing attacks. Based on this analysis, we define the attacker model for these attacks on facial recognition systems. Then, we demonstrate the ability of deepfakes to spoof two commercial facial recognition systems. Finally, we discuss possible means to prevent such spoofing attacks.Deepfakes are media generated by deep learning and are nearly indistinguishable from real content to humans. Deepfakes have seen a significant surge in popularity in recent years. There have been numerous papers discussing their effectiveness in deceiving people. What's equally, if not more concerning, is the potential vulnerability of facial and voice recognition systems to deepfakes. The misuse of deepfakes to spoof automated facial recognition systems can threaten various aspects of our lives, including financial security and access to secure locations. This issue remains largely unexplored. Thus, this paper investigates the technical feasibility of a spoofing attack on facial recognition. Firstly, we perform a threat analysis to understand what facial recognition use cases allow the execution of deepfake spoofing attacks. Based on this analysis, we define the attacker model for these attacks on facial recognition systems. Then, we demonstrate the ability of deepfakes to spoof two commercial facial recognition systems. Finally, we discuss possible means to prevent such spoofing attacks.